nah, it was just peter moore helping his old team out. It happens in business, its all about connections.
Perfectly explained.
:max_bytes(150000):strip_icc()/monopsony.asp-final-b5584ce2bc8c49a19f26b742abd82b8a.png)
Monopsony: Definition, Causes, Objections, and Example
A monopsony is a market condition in which there is only one buyer.www.investopedia.com
Yeah, with a weird hardware design like the PS3's, Sony first party were always going to be the only ones who really utilized the hardware to its fullest potential. With the PS5's hardware quirks, I expect more inpressive games to come out once their top studios leave cross-gen behind as well, though it should be much easier for 3rd parties to get results out of the PS5 than it was for the PS3.
Never question Mark Cerny's genius when it comes to that kind of thing.
Another factor in Series X performance issues is that, despite having more CUs, it doesn’t have a commensurately larger L2 cache, so there can, and probably are, times where CUs are essentially stuck waiting to offload to the cache for a cycle or two, which can lead to various kinds of hiccups, and each hiccup can have a cascade effect through the rest of the system as other processes are knocked back by waiting for data.My limited knowledge of hardware design (very limited) only allows me to speculate, but I believe the reason as to why Cerny went with the smaller APU is related to thermal and power envelopes, and the desire to clock it faster. I suspect that due to the constraints of a console APU, having fewer compute units would allow for a faster clock rate within the same budget. The faster clocks help overall tasks do be performed faster (duh) as long as occupancy remains high.
Talking about occupancy and even memory, there are fundamental design differences between the PS5 and the Series X, architecture wise. The sentence "narrow and fast" and "wide and slow" are applicable here.
The PS5 GPU is as follows
The Xbox Series X however is as follows
- 2 Shader Engines with 2 Shader Arrays each
- Each Shader array has 10CU
- 4MB L2 cache, which I believe is split in half between shader engines
- 2 pairs of CUs are disabled for yields for a total of 36 active CUs
On a first look the Xbox seems to have the advantage, however when you look deeper into the subject, you can tell that their wide approach (more compute units per shader array), as well as the smaller amount of L2 per CU (0.111MB/CU on the PS5 vs 0.096MB/CU on the Series X), and couple it with the slower clock, may prove to be a disadvantage.
- 2 Shader Engines with 2 Shader Arrays each
- Each Shader Array has 14CU
- 5MB of L2 cache with the same conditions as the PS5
- 2 pairs of CUs are disabled for yields, for a total of 52 active CUs
RDNA2 cards kept with the 20 CU per Shader Engine scheme, or 10 per array, so even AMD privileges the more narrow design (here's the floor plan for the Navi21)
The same is applicable to RDNA3 cards, keeping under the 20 CU per Engine / 10 per array
The thing is, people should have never expected the PS5 to win resolution battles. I said this plenty of times, but the raw "grunt" of the Series X should allow it to produce higher resolutions with less "side-effects" when compared to the PS5. The strengths of the later are shown on efficiency: you often have games that perform better / more stable on the PS5, and with a small resolution decrease. Of course engines will prefer different things, so your results may vary, but this is the case more often than not. Additionally, when looking at architectural differences, I believe the Series X cannot achieve those extra frames it sometimes drops simply by targeting resolution, meaning it has an occupancy issue somewhere.
Overall the PS5, from an hardware standpoint, is a better optimised console overall IMO. Additionally, I must add that the "variable clocks" shenanigans was very poorly explained and something people still fail to understand. It was never about "underclocking" the machine, but rather to use a variable power envelope as opposed to a fixed one. A practical example as to why this is, in theory at least, better.
Console A has a fixed power envelope of 200W, with 120W dedicated to the GPU and 80W dedicated to the CPU. Console B has the same fixed power envelope, however it can starve the CPU or GPU of power if the other needs it. The more "math" you run, the more power you need, so ideally you want to have as many transistors firing as possible at once (you never reach 100%, which is your theoretical TFLOP measure). So, how does it happen. When constructing a frame, systems can parallelise their math between CPU and GPU when constructing a frame (long were the times were this was not possible).
Now, if your target is 16ms for a smooth 60 fps, you need both to complete their task under that time, at the same time. Let's now imagine your CPU can construct a frame in 6ms, but your GPU, due to lack of occupancy or power, can only complete said frame in 18ms? You're already below target, and are wasting precious CPU cycles simply waiting for the GPU. When coding for console A, you will either have to reduce the load on GPU, or assume slower performance. On Console B, if the issue is related to power, you can simply direct most of the power from your CPU (which has completed its task) to the GPU (to ensure it constructs the frame in time). These things happens several times per millisecond, and can happen interchangeably.
So, you not only have a console that runs at a higher clockspeed, has more L2 memory per CU and can fill them more easily with work (due to the smaller chain), but you also have a system that intelligently diverts power to where its needed to prevent stalls (assuming your problem is power budget of course).
Apologies for the rant.
Making it smaller has to do with wafer yields and cost per die. Clocking it higher means more difficulty cooling because you have a higher concentration of thermals, which is why they went with liquid metal. the PS5 die is 16% smaller than the series X die. If you plug in the numbers based on TSMC's publicly stated 0.09 defects/mm^2 on 7nm, 3mm edge loss, 0.2mm scribe lanes, a wafer would produce ~195 ps5 dies and ~160 series x dies. now ofc the tapeout for the series wafers is allegedly 3 series s dies per series x dies so the xbox production would be a bit more efficient than that, but on a per wafer cost of $10,000 per 7nm wafer, a ps5 die comes out at around $51 and a series x die comes out at around $62. Then you have a node shrink for ps5s going onto 6nm production... based on rough estimates after a node shrink the ps5 die size went down ~13% which yields 218 ps5 dies per wafer, bringing the cost per ps5 die down to $45-46. That's the kind of efficiency you need if you don't want to take a $100-200 loss on every console.
Too bad none of that will help, as behavioural remedies have already been written off by the CMA.
I've seen people who view this as Microsoft’s chance to verbally lay into everyone protesting the deal in person and trigger a domino effect that leads to the deal being passed with minimal concessions, though that feels like wishful thinking to me.
With the mass of documents MS requested, an extension just to make sure they had everything together and appropriately reviewed and redacted where necessary seems totally in order.
Given the time frame of how late in the year this deal is supposed to go to court with the ftc, there's the risk of it being killed overseas before these documents even matter. I'm just curious as to why resetera are putting so much hope into this being a secret weapon for MS.
I ran google English translate through this and I still don't understand what you are trying to say.My limited knowledge of hardware design (very limited) only allows me to speculate, but I believe the reason as to why Cerny went with the smaller APU is related to thermal and power envelopes, and the desire to clock it faster. I suspect that due to the constraints of a console APU, having fewer compute units would allow for a faster clock rate within the same budget. The faster clocks help overall tasks do be performed faster (duh) as long as occupancy remains high.
Talking about occupancy and even memory, there are fundamental design differences between the PS5 and the Series X, architecture wise. The sentence "narrow and fast" and "wide and slow" are applicable here.
The PS5 GPU is as follows
The Xbox Series X however is as follows
- 2 Shader Engines with 2 Shader Arrays each
- Each Shader array has 10CU
- 4MB L2 cache, which I believe is split in half between shader engines
- 2 pairs of CUs are disabled for yields for a total of 36 active CUs
On a first look the Xbox seems to have the advantage, however when you look deeper into the subject, you can tell that their wide approach (more compute units per shader array), as well as the smaller amount of L2 per CU (0.111MB/CU on the PS5 vs 0.096MB/CU on the Series X), and couple it with the slower clock, may prove to be a disadvantage.
- 2 Shader Engines with 2 Shader Arrays each
- Each Shader Array has 14CU
- 5MB of L2 cache with the same conditions as the PS5
- 2 pairs of CUs are disabled for yields, for a total of 52 active CUs
RDNA2 cards kept with the 20 CU per Shader Engine scheme, or 10 per array, so even AMD privileges the more narrow design (here's the floor plan for the Navi21)
The same is applicable to RDNA3 cards, keeping under the 20 CU per Engine / 10 per array
The thing is, people should have never expected the PS5 to win resolution battles. I said this plenty of times, but the raw "grunt" of the Series X should allow it to produce higher resolutions with less "side-effects" when compared to the PS5. The strengths of the later are shown on efficiency: you often have games that perform better / more stable on the PS5, and with a small resolution decrease. Of course engines will prefer different things, so your results may vary, but this is the case more often than not. Additionally, when looking at architectural differences, I believe the Series X cannot achieve those extra frames it sometimes drops simply by targeting resolution, meaning it has an occupancy issue somewhere.
Overall the PS5, from an hardware standpoint, is a better optimised console overall IMO. Additionally, I must add that the "variable clocks" shenanigans was very poorly explained and something people still fail to understand. It was never about "underclocking" the machine, but rather to use a variable power envelope as opposed to a fixed one. A practical example as to why this is, in theory at least, better.
Console A has a fixed power envelope of 200W, with 120W dedicated to the GPU and 80W dedicated to the CPU. Console B has the same fixed power envelope, however it can starve the CPU or GPU of power if the other needs it. The more "math" you run, the more power you need, so ideally you want to have as many transistors firing as possible at once (you never reach 100%, which is your theoretical TFLOP measure). So, how does it happen. When constructing a frame, systems can parallelise their math between CPU and GPU when constructing a frame (long were the times were this was not possible).
Now, if your target is 16ms for a smooth 60 fps, you need both to complete their task under that time, at the same time. Let's now imagine your CPU can construct a frame in 6ms, but your GPU, due to lack of occupancy or power, can only complete said frame in 18ms? You're already below target, and are wasting precious CPU cycles simply waiting for the GPU. When coding for console A, you will either have to reduce the load on GPU, or assume slower performance. On Console B, if the issue is related to power, you can simply direct most of the power from your CPU (which has completed its task) to the GPU (to ensure it constructs the frame in time). These things happens several times per millisecond, and can happen interchangeably.
So, you not only have a console that runs at a higher clockspeed, has more L2 memory per CU and can fill them more easily with work (due to the smaller chain), but you also have a system that intelligently diverts power to where its needed to prevent stalls (assuming your problem is power budget of course).
Apologies for the rant.
Why the heck is this dude getting gifted an icon sub and not someone like Elder legend, FFS people get your priorities right people.
speaking of is there any way to get rid of the notification at the bottom of every page about the gifted post?Why the heck is this dude getting gifted an icon sub and not someone like Elder legend, FFS people get your priorities right people.
Hugs to you @Yobo
That's a very good question, I have not found a way yet but I will put a single to our superhero mods/admins/sluts
